chat model, answer-letter probabilities · League 2
Gemma 4 31Bon · NVIDIA RTX PRO 6000 Blackwell + NVIDIA DGX Spark
FinishedPublished
Results f24d1f41, updated 16 November 2026, 21:24 UTC; scored by GNU Backgammon 1.08, 2-ply.
How we called it
- Model ID
- gemma-4-31B-it
- Provider or endpoint
- NVIDIA RTX PRO 6000 Blackwell (96 GB), fp8 (vLLM); NVIDIA DGX Spark (GB10) 19 Oct 18:36-22:26 UTC and from 20 Oct 21:10 UTC (back on the NVIDIA RTX PRO 6000 Blackwell (96 GB) 21 Oct 01:25-05:00)
- Date run
- 2026-10-19 to 2026-10-26
- Price
- no price per token; estimate: GPU time, at most; source content/own_cost.json (wall clock on our machines × hourly rate); price date not recorded
- Model type
- chat model, answer-letter probabilities
- Weights
- open
- Runs on
- our GPUs
- Reasoning setting
- thinking on (template default; no effort setting)
- Confidence source
- letter probability
- Rules text
- long
- Option groups
- up to 26 options per request (letters A-Z); more legal plays are split into equal groups, a winner per group, then among the winners; in this data 252 decisions were split into groups of 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26
- Sampling settings
- temperature 1.0
- Notes
- moved between our GPUs and a DGX Spark during the league with the same image, checkpoint and arguments
Results in League 2
- Error rate (PR)
- 19.617.5 to 21.7
- Decisions
- 2,122 (2,122 scored, 0 invalid)
- Engine's best play
- 59.1%
- Blunder rate
- 15.6%
- P95 equity loss
- 0.203
- Mean loss, worst 5%
- 0.348
- Median / P90 s per decision
- 120.6 / 394.3
- $ per 1,000 decisions
- data-v="34.943">34.943estimate · RTX PRO 6000 Blackwell
- Risk judgement, margin 0.001
- 64.5 (95% interval 62.9 to 66.2); risks found 31.9% (223 / 700), right safety 97.1% (878 / 904)
- Risk judgement, margin 0.02
- 65.6 (95% interval 63.8 to 67.5); risks found 34.7% (213 / 613), right safety 96.5% (956 / 991)
- No trade-off
- 518 (left out of every risk metric)
Risk call card
64.5risk judgement, margin 0.001
right risks223
wrong risks23
needless risks26
missed risks454
right safety878
Hit risk only: being hit. Gammon risk and priming are not in it.
Close choices
A decision is a close choice when the second most likely play has at least half the probability of the first (internally: split).
Probabilities over plays, not sampled reasoning. Method
League 2: complete letter-readout records are mostly few-option positions, a selected subset.
- Decisions
- 2,122
- Eligible (full probability distribution)
- 169
- Left out, by reason
- asked in groups of options 175; incomplete distribution (an option without its own probability) 1,179; not audited 599
- Close choices
- 0.0% (0 / 169), under 200 decisions
- Ties in the top two
- 0 decisions (0 at a positive probability, 0 at 0; the two most likely plays tied: 0, the second tied with the third: 0); 0 of them close choices. Ties are broken by the order the plays were shown, never by the engine.
- Between a safe play and a riskier one
- N/A (0 / 0) of the close choices; 0.0% (0 / 169), under 200 decisions of all eligible decisions
| Decisions | Number | p1 (most likely play) | p2 (second) | p1 + p2 |
|---|---|---|---|---|
| Close choices | 0 | N/A | N/A | N/A |
| Other decisions | 169 | 1.000median 1.000 | 0.000median 0.000 | 1.000median 1.000 |
| p1 + p2 | 0.0–0.1 | 0.1–0.2 | 0.2–0.3 | 0.3–0.4 | 0.4–0.5 | 0.5–0.6 | 0.6–0.7 | 0.7–0.8 | 0.8–0.9 | 0.9–1.0 |
|---|---|---|---|---|---|---|---|---|---|---|
| Close choices | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| Other decisions | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 169 |
| Decisions | Number | Hitting rolls | Newly made points | Pip counts | Longest prime | None / 1 / 2 / 3 / 4 differ |
|---|---|---|---|---|---|---|
| Close choices | 0 | N/A0 / 0 | N/A0 / 0 | N/A0 / 0 | N/A0 / 0 | 0 / 0 / 0 / 0 / 0 |
| Other decisions | 169 | 40.2%68 / 169, under 200 decisions | 10.7%18 / 169, under 200 decisions | 17.2%29 / 169, under 200 decisions | 28.4%48 / 169, under 200 decisions | 53 / 73 / 39 / 4 / 0 |
| Decisions | Number | Gap between the two plays' values | Most likely play: behind the engine's best | Second most likely play: behind the engine's best |
|---|---|---|---|---|
| Close choices | 0 | N/A | N/A | N/A |
| Other decisions | 169 | 0.142median 0.054 | 0.008median 0.000 | 0.137median 0.045 |
| Margin | In Risk pays positions | In Safety pays positions |
|---|---|---|
| margin 0.001 | N/A0 / 0 | N/A0 / 0 |
| margin 0.02 | N/A0 / 0 | N/A0 / 0 |
| Margin | Risk pays | Safety pays | No trade-off | Risk pays, small stake | Risk pays, medium stake | Risk pays, big stake |
|---|---|---|---|---|---|---|
| margin 0.001 | 0.0%0 / 14, under 200 decisions | 0.0%0 / 57, under 200 decisions | 0.0%0 / 98, under 200 decisions | 0.0%0 / 4, under 200 decisions | 0.0%0 / 8, under 200 decisions | 0.0%0 / 2, under 200 decisions |
| margin 0.02 | 0.0%0 / 10, under 200 decisions | 0.0%0 / 61, under 200 decisions | 0.0%0 / 98, under 200 decisions | N/A0 / 0 | 0.0%0 / 8, under 200 decisions | 0.0%0 / 2, under 200 decisions |
At the 0.02 margin a Risk pays position has a stake of at least 0.02, so the small band is empty. A big stake means playing safe costs more by the engine's value, not that the decision was harder.
| Margin | Close choices: missed risks | Other decisions: missed risks (weighted to the close choices' mix) | Difference, close minus other | Crude difference |
|---|---|---|---|---|
| margin 0.00113 decisions | N/AN/A (0 / 0) | 76.9%76.9% (10 / 13), under 200 decisions | N/A: no close choices in this group | N/A |
| margin 0.029 decisions | N/AN/A (0 / 0) | 77.8%77.8% (7 / 9), under 200 decisions | N/A: no close choices in this group | N/A |
By game phase
| Phase | Error rate (PR) | Decisions |
|---|---|---|
| Opening | 19.0 | 298 |
| Contact | 24.3 | 1,367 |
| Holding game | Too few decisions | 65 |
| Race | 6.0 | 228 |
| Bear-off | Too few decisions | 164 |